Quiz collegato alla guida · Medio Livello
Quiz su alimentazione e raffreddamento nei data center AI
Spiegare l'erogazione di potenza del data center, la rimozione del calore, il raffreddamento a liquido diretto, i compromessi PUE e acqua per i sistemi di accelerazione IA densi.
Domanda 1 di 8
Perché i rack di acceleratori AI possono richiedere design di raffreddamento diversi da quelli dei rack di server a bassa densità?
Anteprima di ogni domanda8 domande
Le risposte e le spiegazioni vengono visualizzate mentre rispondi al quiz.
- Perché i rack di acceleratori AI possono richiedere design di raffreddamento diversi da quelli dei rack di server a bassa densità?
- Because accelerators produce less heat than most storage equipment
- Because rack dimensions eliminate the facility heat load
- They can concentrate more electrical power and heat in a smaller space
- Because accelerator racks do not need electrical distribution
- Cosa fa il raffreddamento liquido diretto al chip?
- Circulates coolant through the entire server room instead of using air
- Circulates coolant near hot components to carry heat into a facility loop
- Removes all heat without a heat-rejection system
- Exhausts server air directly outside without heat exchange
- Cosa confronta il PUE?
- IT equipment energy divided by total facility energy
- Water consumed per IT energy unit
- Compute output per rack area
- Total facility energy divided by IT equipment energy
- Cosa non misura il PUE da solo?
- Water consumption or carbon intensity
- Total facility energy
- IT equipment energy
- The energy ratio between facility and IT equipment
- Perché i progetti di raffreddamento possono scambiare elettricità con l'uso dell'acqua?
- Lower cooling energy always means lower water use in every climate
- Liquid cooling eliminates all facility water requirements
- Cooling methods such as evaporation have different energy and water requirements
- Power draw has no relationship to heat removal
- Which examples are part of a data center’s cooling load?
- GPU compute cores and memory banks
- Fans, pumps, chillers, and cooling towers
- Training jobs and optimizer states
- Storage systems and network interfaces
- What monitoring can help detect liquid-cooling risks?
- Model validation accuracy and label quality
- Batch size and data split consistency
- The number of cloud regions used for training
- Coolant flow, pressure, temperature, and leak sensors
- Why should facility planning use measured workload profiles?
- Measured power and utilization can differ from nameplate maximums
- GPU specifications fully determine site impact
- A single benchmark predicts facility demand for every workload
- Local climate has no effect on cooling design